"Chat GPT will be soon integrated into Windows with Copilot" this is the lead from the course I took on June 28th under the aegis of the International Science Council, with Nick Scott ISC senior consultant and Zhenya Tsoy as ISC conductor. We have been learning on the use of ChatGPT for science. Below you find my notes.
"Since the launch of ChatGPT in November 2022, the hype around generative AI (artificial intelligence) has been immense. Products that can automatically generate text, images and video when prompted to do so are being discussed as a game-changer for organizations and society at large. Generative AI promises organizations and their staff a route to increase their productivity, supporting them in automating – or at least delivering more quickly and easily – many important regular tasks. At the request of ISC Members, ISC has organized a practical workshop to create a safe space for experimentation and explore how the most popular tool – ChatGPT – can be used in our everyday work. The ISC senior consultant Nick Scott has shared insights on how to work around technical capabilities and limitations of the tool, what makes for an effective prompt, and answered practical questions around the use of ChatGPT."
An individual language learner (LLM model)
Chat GPT is more capable of being an assistant for us in our work.
Recommendations
- Give Chat GPT a role: roles give Chat GPT what role to play: "Hellow, Act as a consultant and trainer, with a specialism in digital and AI"
- Give ChatGPT CONTEXT: content and detailed background información, such as background info.
- I can tell CONVERSE, chat and respond in order to make sure that it understands. Tell it the strong and weak points in the answer.
- FOCUS: on the areas of your expertise. If it is not in your area, it will be difficult to see true.
- POLITE: gets better results
- GUIDE: Tell what you want
- SEPARATE chat by topic
Strengths
- LATERAL THINKING is a strength of ChatGPT. This includes videos for instance, not only texts, poems, tables, telling stories in different ways for different audiences or in the way of a particular person/politician.
- AI, generative AI as a subset of AI, Large Language models, GPT (an specific language model)
Suggestions
- Breaking a task into two tasks or steeps: first making a text more understandable "Please rewrite this text in simple understandable plain english, readable to a ..."
- Help with preparations for a performance review meeting, -pay rise- with a boss.
Limitations and risks
- Bias content, ie. more men
- Copyright, that may be a concern with info consumed and produced
- Misuse for misinformation and disinformation. The model can create both, it can create to deceive
- Cut out of data since 2021, no Internet access, Eager to please so, it makes stuff up.
- Data privacy, in the chat history and training button should anonymize for personal information. Be very careful of what you put, it will consume it unless you have the button off
- You will not be getting the same answer twice
- Free version in often inaccessible
- Limited memory
No comments:
Post a Comment